The vendor nobody vetted
In 2026, some of the year's most damaging corporate breaches have traced not to the company under attack but to a small, low-visibility vendor several tiers removed from it. In September, Canva confirmed that enterprise customer contracts, invoices, and correspondence had been exposed — not through its own systems, but through Canny, the roughly 20-person feedback-widget vendor it used to collect product suggestions. Attackers spent two weeks inside Canny's infrastructure, which held live admin-level connections into Canva's Salesforce CRM, Jira issue tracker, Okta identity provider, and Zendesk support desk. By their own account, the same access still reached over 300 other companies' Jira instances, including 30 in the Fortune 500.
A few weeks earlier, Trezor hardware wallet customers ran into the same problem from the opposite direction — not too much access, but a promise that didn't hold. Trezor's fulfillment partner, ShipMonk had told customers in writing that order data was deleted under a 90-day retention policy. A SQL-injection breach in August proved otherwise: investigators found shipping records dating back to 2019 still sitting in ShipMonk's systems, and the disclosed customer count grew nearly sixfold — from just under 14,000 to roughly 81,000 — once the retained data surfaced.
They weren't isolated, either. Ledger, another hardware wallet maker, disclosed its own breach the same year, after an intrusion at Global-e — the partner that runs its online store's order processing — exposed customer names, addresses, and phone numbers. It was the second time in six years a partner sitting between Ledger and its customers, not Ledger itself, has been the point of failure.
In every case, the named company's own security held. They were all taken down by someone smaller, standing a step or two back in the chain, that nobody was really watching.
It's always the vendor you didn't vet
Most companies review the vendors on their contract list and call it done. But the actual risk usually sits one layer past that — the tool your vendor plugged into their CRM, the sub-contractor running their fulfillment, the scrappy startup they quietly depend on. Nobody signs anything with those companies. Nobody reviews them. And they can end up with just as much access as anyone you did vet: SecurityScorecard's research puts the ratio at 60 to 90 indirect, never-contracted fourth-party relationships for every vendor a company actually reviews — so a program managing a disciplined 80 direct vendors is still sitting on somewhere between 4,800 and 7,200 it's never looked at.
It's not just that nobody's looking — it's that almost nobody can. Panorays' 2025 survey of 200 CISOs found only 3% have full visibility into their supply chain past the first tier. Which tracks: Canny wasn't hiding anything from Canva, it just was never anyone's job at either company to ask what a 20-person feedback tool could actually reach once it was wired into Salesforce.
AI agents make it worse, faster
Smaller companies are shipping AI agents quickly — fewer people, more to automate — without the security team to watch what those agents can touch. And this isn't just a small-company problem: OneTrust's 2026 AI-Ready Governance survey found 87% of organizations encourage AI agent use, while only 47% say they have clear governance, oversight, and controls to match. If the well-funded end of the market can't keep up, it's a safe bet a lot of the smaller vendors quietly plugged into your stack can't either.
It's the same story everywhere you look. Deloitte's 2026 State of AI in the Enterprise survey of 3,235 leaders found only one in five companies has a mature model for governing autonomous AI agents; IBM's research puts the share with a current, complete inventory of their own agents at just 18%, and finds 68% of executives already worried their AI initiatives will fail for lack of the integration work that inventory would require. If the big, well-resourced companies are struggling to keep track of their own agents, a small vendor two tiers down your supply chain almost certainly isn't doing better.
What two real breaches have in common
Canva and Trezor failed in opposite-looking ways — one had too much access sitting with a vendor nobody was watching, the other trusted a written promise nobody checked — but the root cause was identical: a small partner had reach, or held data, long after anyone at the parent company had a reason to still be thinking about it. Canny's Salesforce connection was presumably granted for a legitimate reason once; ShipMonk's 90-day deletion policy was presumably true when it was written. Neither company revisited either fact until an attacker did it for them, and in both cases the gap had been sitting open for years before it was found.
The contract probably doesn't cover this either
Even if you wanted to hold that vendor's vendor accountable, most contracts weren't built for it — sub-processor clauses usually assume one layer down, not three. Regulation mostly doesn't help either: NIS2 sets its toughest requirements, and its proactive audits, for medium and large entities — 50 or more employees, and €10M or more in turnover or balance sheet — while explicitly exempting micro and small enterprises. Canny, at roughly 20 people, would likely fall under that line. So the accountability ends up sitting with whoever's easiest to name, not whoever actually failed.
What actually helps
Ask what your vendor's vendors can touch, not just what the vendor itself can touch — Canny's access into Canva's CRM was the whole incident.
Audit the admin-level connections your vendors hold, into your CRM, identity provider, and support desk specifically — those are exactly the systems attackers went after at Canny.
Don't file "we deleted it" and move on — a written assurance is a claim, not proof. Check back, the way Trezor eventually had to.
Review a small AI tool the same way you'd review a vendor, regardless of how small the company behind it is.
Know your own blast radius. If a vendor plugged into your CRM got popped tomorrow, could you say quickly what it could actually see? If not, that's the gap worth closing first.
None of this is complicated, it just isn't the review most companies are set up to do — especially once you're looking three tiers down instead of one, or auditing what a vendor's own connections can reach. That's basically the gap Granur closes: one place to see not just your vendors, but what's actually connected to them, agents included, so a question like this doesn't have to wait for a breach notice — or a headline about your feedback widget — to get asked.